Best Practices, Heroes

10 Tips for Choosing a Mobile Demand-Side Platform

Stara Gipson and Brad Puder by Stara Gipson and Brad Puder | July 9, 2018

Brad Puder is a Performance Marketing Manager for War Dragons at Pocket Gems. He worked for a variety of startups in Chicago before joining the Pocket Gems’ product team, where he worked on Live Operations for War Dragons. He then pivoted to focus on growth for the game and now leads programmatic buying for Pocket Gem’s Core Studio.

Learn more from his Mobile Hero profile.

Programmatic buying is all the rage these days. The biggest upside of course is the near limitless amount of scale. Billions of devices across the world are made accessible via hundreds of exchanges and SSPs, and working with the right Demand-Side Platform (DSP) can get you access to the kind of traffic you’re looking for.

Historically, programmatic has been tough on mobile. It’s been a volatile landscape, quality has been questionable, brands have distorted prices, and technological constraints have all made real-time bidding challenging, especially for advertisers who have been able to maintain scale with big social channels and some more traditional ad networks. The times are changing though.

There are a lot of DSPs out there, and new DSPs are popping up faster than networks. There is a ton of overlap in terms of offerings, and on paper they should all very much behave the same, but there are a number of factors to consider when adding a DSP to your channel portfolio.

Why now is the right time to invest in programmatic

The prevalence of waterfall mediation historically meant that programmatic buyers typically had access to lower quality traffic as the result of publishers granting exchanges/SSPs lower priority. The death of the waterfall has been slow on mobile, but it’s definitely ramping up in 2018. There’s no true “header bidding” on mobile, but Unity’s true unified auction, parallel bidding with Max, open bidding at Google, advanced bidding on MoPub, super auctions with Oath, Fyber’s FairBid, etc. are all paving the way towards more unified auctions, effectively giving programmatic buyers access to the same impressions as networks. We are some time away from this happening, but the transition is happening fast.

Additionally, networks are making their inventory available programmatically as well, which means that DSPs can access network inventory by leveraging their own algorithm and bidding model. In general, this means the industry is beginning to converge, and that supply path is becoming more transparent.

1. Pick a partner with a business model that works for you

Different partners are able to offer buying on CPM, CPC, CPI or CPA. This should be part of your decision-making process. Buying on CPM gives you the most flexibility and transparency on programmatic, but unless you have a solid optimization method, this can be tricky given the volatility of RTB. When you are buying on CPM, it can often take longer to see optimal CPI and CPA results. Also bear in mind that some partners may offer a CPM Plus model, where you pay a fixed fee on top of your spend, while others may push for a non-transparent margin on CPM buys, similar to CPI. Buying on CPI and CPA de-risk you in the short-run, but DSPs also have to de-risk the opportunity for themselves by taking a substantial, non-transparent margin from the expected CPMs from your CPIs. In other words, in order to make CPI work for both parties, a certain degree of arbitrage happens. Some DSPs might have other varying fees for ad serving or creative service, so at the end of the day, you should evaluate whether the cost structure works for your UA strategy and risk appetite.

2. Make sure your DSP is getting you access to the right traffic

Almost every DSP is going to tell you they have access to the big mobile exchanges, like MoPub and Doubleclick Exchange (Google AdX). Some access exchanges and SSPs might be a better fit for your UA needs (like network inventory or more gaming-specific inventory). Some DSPs buy in-app only. Some DSPs buy more banner and native traffic than others. And others are big on rich media/playables or other emerging ad unit types, like native video. When considering a DSP, be upfront about the kinds of traffic you’re looking for. Quantify your risk appetite by indicating what percentage of your budget you want to allocate towards more sustainable, historically-proven methods of growth versus more experimental methods.

DSPs are constantly rolling out new features and products, so staying up to date with their product roadmap will help you think about testability plans.

3. Your DSP should be transparent

DSPs buying from open auctions should be able to share some of the common data points available from OpenRTB requests, namely the publisher name, app store ID, or bundle in addition to the exchange name. This helps you to evaluate whether the DSP is acquiring inventory from the right sources, alleviate some forms of fraud, and ensure that the DSP is providing real, expected traffic and not unexpected affiliate traffic. Be wary of DSPs who claim that essential, widely available information from OpenRTB or the programmatic ecosystem is “proprietary data.”

4. Evaluate your DSP’s bidding model and optimization strategies

From an algorithm perspective, network and social channels have an “invisible hand” in their marketplaces. Their algorithms cover both demand and supply-side, and the market gravitates towards generating the highest possible yields (eCPMs and impression volumes). For programmatic it’s a different story. The DSPs you work with have robust algorithms and bidding models that only optimize on the demand-side. In other words, DSPs only control one side of the equation, namely what impressions to bid on and how much to bid. This also means that if you are optimizing towards CPI/CPA, DSPs need to carve out margins between the actual CPMs you can afford with your CPI bid and the CPMs they pay in the market.

In general, most DSPs have some sort of user graph. A DSP will listen to over a billion devices in a given month through bid requests form exchanges. These OpenRTB requests contain 40+ parameters from device data and location data to publisher info and sometimes even demographic data. By listening to enough data on a device ID, you can start to build a profile on that device. That profile can be further augmented with first, second, or third-party data like external audiences, advertiser-supplied segments, etc.

Ultimately, DSPs’ algorithms strive to weigh the right combination of data from these parameters as well as your KPIs to inform how much they bid. Though these principles are shared across DSP algorithms, bidding models can be very different. It’s possible that some algorithms are better at some kinds of traffic, or with some industries or verticals. You should make it a priority to understand your DSP partner’s bidding strategy, and if it aligns with your goals.

The other large component of a DSP’s algorithm is how they optimize towards your KPIs. Your DSP should be capable of ingesting post-install data. It’s 2018, and if they can’t, you should really be questioning why. Not only does a post-install event allow the DSP to acquire users more likely to complete a down-funnel action, but it is also an additional performance benchmark that you and your DSP can use to measure performance. Ultimately, a good algorithm should be able to increase the incidence rate of the event that you are optimizing towards, assuming you are feeding it enough data.

Ask your DSP about how they can optimize towards these post-install goals, whether it’s a target conversion rate, or (in many cases even better) a target eCPA. Providing an eCPA goal, even if you’re buying on CPI or CPM, allows your DSP to help ensure they’re bidding the right amount for an impression. If, based on a device’s data in an OpenRTB request, you can assume that the user has a higher likelihood of making a purchase, you wouldn’t want to bid with a CPM computed for an install/CPI goal. Ask your potential partners about how they leverage post-install and eCPA optimization.

Finally, the other big component of a solid algorithm is creative optimization. We all know a strong creative can make a big difference in your install rates, which can be hugely impactful in raising your eCPMs and effectively give you more bidding power in the market. A strong DSP will have a suite of creative optimization tools that ensures you’re always testing and iterating on your creatives to achieve your highest possible eCPMs. There’s two big components of install rate optimization: finding the right audience that your creative is most likely to resonate with, and rendering the strongest possible creative for your given audience. The first is solved for by strategic targeting, while the second is solved for through creative optimization. Many DSPs offer additional creative services, often at no cost, to help you continue to build, refresh, test, and launch creatives. You should ask about how your DSP leverages creative optimization across ad formats they run.

5. Understand how your partner combats fraud

Fraud is prevalent in programmatic. It’s often not as sketchy as network traffic, but depending on what inventory sources you’re buying from, you may be exposing yourself to risk. Every DSP will claim to have a robust fraud solution. You should have an open dialogue with potential partners about how they combat fraud, ensure viewability, and whether or not they’re amenable towards certain anti-fraud incentives in your agreement, like minimum click-to-install time. Again, per the above point, transparency on what supply you’re buying and how also goes a long way. Work with your MMP to identify if the traffic you’re seeing is suspicious.

6. Think through your retargeting strategy

Retargeting and re-engagement are the bread and butter of many DSPs. If you think about it, retargeting is a great use case for programmatic. There’s a lot of noise in UA trying to identify valuable users within billions of devices, but the equation is reversed on the retargeting front. You already have the users you want, and can go out and find them at the most affordable prices you can across corresponding bid requests. This is also the case for leveraging third or second party audiences you’ve obtained and want to acquire programmatically for UA purposes. If you are serious about re-engagement, identify how your DSP can find the audiences you’re looking for.

7. Verify your partner’s integrations

Ensure a clean relationship between your attribution partner and your DSP. The tighter the integration, the better. Of course, new DSPs are always popping up and might not be fully integrated, but a full integration allows for seamless post-install data ingestion, real-time suppression, sending non-attributed data, fraud detection, etc.

If you’re looking at users across devices, consider what cross-device solutions your DSP is leveraging. Similarly, if identity resolution is key to your marketing strategy, you should understand what tools your DSP has at its disposal. If you use a data management platform (DMP) and you have data that could be optimal for your programmatic strategy, ensure there’s a relationship between your DSP and DMP.

8. Make sure you’re getting the attention you deserve

It goes without saying that you should seek out DSPs with strong customer success teams. You want an account manager who’s committed to your success, and can provide you with the advice, resources, and tools you need to grow. Having an account manager that has a more technical understanding of the DSP’s bidding model is also useful. DSPs that foster strong cross-functional relationships between account management and product/data teams are ultimately going to provide you with a greater level of insight and strategy than those with account managers who are more removed from the technical side.

9. Understand how much control you will have

On the DSP front, there’s a self-serve managed spectrum. On one end of the spectrum, there are DSPs who are entirely self-serve, and really just provide you with raw access to exchange inventory. These are the most flexible, but require the most amount of resources, work, and money to set up. On the opposite end of the spectrum, you have fully managed services, which entirely leverage a DSP’s proprietary data and bidding models, obfuscate a lot of transactional and inventory data form the buyer, and are tasked with most of the campaign optimizations. Managed services generally have more robust algorithms and of course incur a higher margin.

Most DSPs fall somewhere in the middle of this spectrum, offering buyers some level of control and customization, but you should assess what your programmatic needs and risk appetite are in both the short and long-run. Can you afford to take on huge losses and heavy risk as you experiment towards a more customized model for yourself? Or would you be better off leveraging a managed service’s robust user graph for campaign optimization? Have a candid conversation with your DSP about your goals, bandwidth, and what level of optimization you would prefer to be taking on yourself, leveraging your own first-party data (e.g. pub-level bidding), versus what you’re comfortable to completely pass on to the managed DSP. If you’re considering working with multiple DSPs, you might want to consider balancing self-serve and managed offerings.

10. Evaluate how you work with multiple DSP partners

If you know programmatic is the right path for you, you should be definitely exploring multiple DSPs. Theoretically, in a world in which there is information asymmetry and equal inventory access, you would find one DSP whose model works best for you and pump all your resources behind it. But there are a number of differentiating factors across DSPs (just read the points above), and enough that you may want to consider working with a portfolio of DSPs, at least in the short to medium run. Think about how you’d want to diversify this portfolio based on the strengths of each DSP.

As you allocate budgets across DSPs, try and balance factors like creative types, inventory sources, managed versus self-serve, customer-base (does this DSP work with primarily e-commerce or gaming customers), versus your marketing needs. Of course, try and ensure you are still balancing your sustainable growth versus marketing R&D/experimentation ratio across this portfolio and not just running the same type of campaign and targeting across all DSPs or running a bunch of radical experiments (unless you can afford to do so).

Ultimately, a DSP’s model becomes more effective for you the more data it ingests on your behalf, so don’t spread yourself too thin by preventing a single DSP from learning effectively. That being said, once you’re fully ramped up with multiple DSPs, you should absolutely be looking at what partners are able to consistently hit KPIs like eCPA goals.

Some exchanges even provide services directly to advertisers. MoPub, for example allows large buyers to understand how effective their media buys are across DSPs they work with, so you have an understanding of what your data looks like on the supply-side, agnostic of which/how many DSPs you are buying from.


There is a lot to consider when working with a new DSP partner. You should consider all of the points above, but as a cheat sheet here are some key questions to ask your DSP before you start buying:

  • How are you different from your competitors?
  • What companies similar to mine do you work with?
  • What exchanges do you work with?
  • What type of ad formats do you run?
  • How do you handle creative optimization?
  • How does your algorithm work?
  • Can I buy via CPM, CPI, or CPA? And how much data is required before I can transition to CPA?
  • How do you handle post-install optimization? Can I provide you with target eCPAs?
  • How are you combating fraud?
  • What information can you pass back to me via macros?